Referencing Strategy for the Direct Comparison ... - ACS Publications

Department of Chemistry, University of California, Irvine, California 92697. J. Phys. Chem. B , 2010, 114 (2), pp 929–939. DOI: 10.1021/jp905286h. P...
0 downloads 0 Views 3MB Size
J. Phys. Chem. B 2010, 114, 929–939

929

Referencing Strategy for the Direct Comparison of Nuclear Magnetic Resonance and Molecular Dynamics Motional Parameters in RNA Catherine Musselman† and Qi Zhang‡ Department of Chemistry, UniVersity of Michigan, Ann Arbor, Michigan 48109

Hashim Al-Hashimi§ Department of Chemistry and The Biophysics Program, UniVersity of Michigan, Ann Arbor, Michigan 48109

Ioan Andricioaei* Department of Chemistry, UniVersity of California, IrVine, California 92697 ReceiVed: June 4, 2009; ReVised Manuscript ReceiVed: September 18, 2009

Nuclear magnetic resonance (NMR) spectroscopy and molecular dynamics (MD) simulations are both techniques that can be used to characterize the structural dynamics of biomolecules and their underlying time scales. Comparison of relaxation parameters obtained through each methodology allows for cross validation of techniques and for complementarity in the analysis of dynamics. Here we present a combined NMR/MD study of the dynamics of HIV-1 transactivation response (TAR) RNA. We compute relaxation constants (R1, R2, and NOE) and model-free parameters (S2 and τ) from a 65 ns molecular dynamics (MD) trajectory and compare them with the respective parameters measured in a domain-elongation NMR experiment. Using the elongated domain as the frame of reference for all computed parameters allows for a direct comparison between experiment and simulation. We see good agreement for many parameters and gain further insight into the nature of the local and global dynamics of TAR, which are found to be quite complex, spanning multiple time scales. For the few cases where agreement is poor, comparison of the dynamical parameters provides insight into the limits of each technique. We suggest a frequency-matching procedure that yields an upper bound for the time scale of dynamics to which the NMR relaxation experiment is sensitive. I. Introduction In recent years it has become clear that describing not only the structure, but also the dynamics of a biomolecule is essential to its full characterization.1-5 This is especially important for RNA, for which conformational flexibility is nearly universally observed to be essential for its binding and function.1,2,4,6,7 However, obtaining a detailed characterization of the dynamics of RNA can be difficult. Nuclear magnetic resonance (NMR) spectroscopy is one of the most powerful methods for the detection of internal motions, with atomic resolution and sensitivity anywhere from the picosecond to millisecond time scales for molecules in the solution state.7-9 However, collecting data on RNA and interpreting dynamical information is not always straightforward, due, in part, to the possible breakdown in assumptions underlying the methods of data interpretation, namely the assumption that the overall and internal motions are decoupled.10,11 Molecular dynamics (MD) simulations12 also provide atomic level detail of solvated molecules and can, in principle, provide a more detailed analysis of structure and dynamics than is possible by NMR. However MD is limited in time scale by the length of the trajectory that can be collected, which can often lead to incomplete sampling of conformational dynamics (i.e., to the so-called broken ergodicity problem13). * Corresponding author. E-mail: [email protected]. † Present address: Department of Pharmacology, University of Colorado Health Sciences Center, Denver, CO. ‡ Present address: Department of Chemistry, University of California, Los Angeles, CA. § E-mail: [email protected].

Additionally, there are inaccuracies inherent to the approximations in the empirical force fields employed that can lead to errors in the simulated dynamics. The comparison of biomolecular dynamics, as derived experimentally by NMR and theoretically by MD simulations, thus provides an important means for cross-validating and interpreting results from each methodology.58-64,67-69 Moreover, these techniques are highly complementary, and thus their combined use provides the basis for a synergistic approach to the characterization of dynamics. Comparison of dynamics between MD and NMR is most often made through NMR spin relaxation parameters.14 This is relatively straightforward when molecular motions are not coupled to overall rotational diffusion. In such a case, the NMR data can readily be interpreted in terms of internal and overall motions.10,11 Further, although overall rotational tumbling is usually not sufficiently sampled during the finite trajectory of an MD simulation, it can generally be removed through the overlay of each trajectory snapshot onto a reference structure,15 allowing for the accurate calculation of internal dynamical parameters. Comparing results from NMR and MD is significantly more complicated when internal motions are coupled to overall rotational diffusion. Such motional correlations have been detected in some proteins16-19 and are prevalent in RNA molecules.20-24 In such cases, parameters describing internal motions are difficult to obtain experimentally because the contributions to relaxation arising from internal motions cannot be easily separated from those due to overall rotational diffusion. Moreover, from a MD simulation standpoint, when these

10.1021/jp905286h  2010 American Chemical Society Published on Web 12/30/2009

930

J. Phys. Chem. B, Vol. 114, No. 2, 2010

Musselman et al.

Figure 1. (a) Elongated construct transcribed for the reported NMR experiment. Added residues are shown in gray and isotopically labeled residues are shown in red. (b) Wild-type HIV-1 transactivation response (TAR) RNA element used in the molecular dynamics simulation. (c) Instead of using a mean reference frame to calculate the relaxation parameters (left), a domain-anchored reference frame (right) is used in order to match the elongated TAR reference frame. (d) HSQC overlay of the nonelongated (red) and elongated (blue) TAR constructs.

motions widely alter the overall shape of the molecule on the time scale of an MD trajectory, the moment of inertia changes significantly and the overall structure can no longer be used as a single reference frame for eliminating overall rotational diffusion. Although this problem has been nicely addressed in the iRED analysis presented by Prompers and Bruschweiler, this method does not allow for accurate computation of model free parameters (the most commonly compared NMR parameter) if motions are coupled. A recently described domain-elongation NMR experiment addresses the problem of motional coupling specifically for RNA.21 In this experiment, the coupling of internal and overall motions is largely eliminated through a substantial elongation of one of the helical domains. This experimental strategy slows overall tumbling and causes it to be dominated by the elongated domain, thus decoupling internal motions that were on the time scale of the overall tumbling of the nonelongated RNA and effectively anchoring the reference frame onto the elongated domain. Here, we mimic this reference frame in the analysis of an MD trajectory by overlaying each trajectory snapshot such that it aligns with the elongated domain, which serves as the reference, allowing us to compute relaxation parameters that can be directly compared to those obtained in a domainelongation NMR experiment. Using this strategy, we examine the dynamics of the wild-type transactivation response (TAR) RNA element of HIV-1 (see Figure 1) for which we have previously shown that the internal and overall motions are inseparable.24 We find good agreement between many parameters, which reveal a picture of complex dynamics in the RNA spanning multiple time scales. II. Theory and Methods A. Relaxation Parameters. NMR relaxation parameters report on the overall and internal motions of a molecule. When cross-correlation effects are suppressed, imino 15N relaxation is dominated by dipolar interactions with the directly bonded

1 H and by the chemical shift anisotropy (CSA) of the 15N. Assuming an exchange term, Rex, of zero and an axially symmetric CSA tensor with the principal axis colinear with the NH bond, one can express the longitudinal and transverse relaxation constants (R1 and R2) and the nuclear Overhauser enhancement parameter (NOE) as follows:25

R1 )

R2 )

d2 (3J(ωN) + J(ωH - ωN) + 6J(ωH + ωN)) + 4 c2 J(ωN) (1) 3 d2 (4J(0) + 3J(ωN) + J(ωH - ωN) + 6J(ωH)) + 8 c2 (4J(0) + 3J(ωN)) (2) 18

NOE ) 1 +

d2γH (6J(ωH + ωN) - J(ωH - ωN)) 4R1γN

(3) Here, J(ω) is the spectral density function, ωN and ωH are the Larmor frequencies in radians per second, d ) (µ0hγHγN/8π2〈r3〉) and c ) ∆σωN, with µ0 as the permeability of free space, h as Planck’s constant, γH and γN as the gyromagnetic ratios of 1H and 15N, respectively, r as the NH bond length, and ∆σ as the CSA of 15N. Relaxation parameters are obtained directly from an NMR experiment and can also be computed from the coordinate trajectory of an MD simulation by calculating spectral density functions from the internal and overall correlation functions. For calculations from a finite molecular trajectory, J(ω) is approximated as

J. Phys. Chem. B, Vol. 114, No. 2, 2010 931

J(ω) ≈ 2

∫0t

max

Ci(t)Caxial o (t) cos(ωt) dt + 2

∫t ∞

max

Ci(∞)Caxial o (t) cos(ωt) dt (4)

Here Coaxial(t) is the correlation function for overall tumbling when modeled with an axially symmetric diffusion tensor,26 Ci(t) is the correlation function for internal motions, and Ci(∞) is the converged plateau value of Ci(t). The first term in eq 4 is computed directly from the MD trajectory with tmax being the maximum time of the calculated correlation functions. The latter term, approximating the unsampled overall tumbling, is calculated from analytical evaluation of the integral. When calculating correlation functions from a molecular dynamics simulation of a flexible multidomain molecule, it is important to define the proper reference frame. Overlaying the atoms of a stable molecular domain across all trajectory snapshots will eliminate all global motions for that domain and will yield correlation functions for bond vectors in that domain which contain information only on the local motions. However, with the same reference frame, bond vectors for another domain in the same molecule will yield correlation functions which contain information on both the local motions as well as motions induced in the bond vector by any interdomain fluctuations. B. Model-Free Parameters. The most popular method for interpreting relaxation data in terms of dynamics is the Lipari-Szabo model-free approach,27 in which the dynamics of bond vectors are described by amplitudes (order parameter, S2) and corresponding correlation times (τ) as obtained through the parametrization of the correlation function. For simple motions, a single exponential fit is used; however for bond vectors undergoing more complex motion, an extended twoexponential form is used,11

Ci(t) ) S2 + (1 - Sf2)e-t/τf + (Sf2 - S2)e-t/τs

(5)

where S2 ) Sf2Ss2 is the tail value of the time correlation function and the f and s subscripts denote “fast” and “slow” motions, respectively. It is assumed that internal and overall motions are not correlated and that the fast and slow internal motions are also not correlated. Model free parameters are obtained from an MD simulation through fitting of the computed internal correlation function, Ci(t). The order parameter, S2, is either obtained as the plateau of the correlation function or is calculated from an equilibrium expression29 given as

Seq2 )

〈1/r3〉2 3 2 2 (〈µˆ x 〉 + 〈µˆ y2〉2 + 〈µˆ z2〉2) + (〈µˆ xµˆ y〉2 + 〈1/r6〉 2 1 〈µˆ zµˆ x〉2) (6) 2

[

]

The corresponding correlation time, τ, is generally obtained from fitting an exponential function to Ci(t) where τ is the timeconstant from the fitting. For bond vectors undergoing more complicated motions, a sum of exponentials may be used, providing a series of correlation time constants. C. Experimental Relaxation Parameters. Two samples of elongated wild-type HIV-1 TAR (see Figure 1b) were prepared by in Vitro transcription, using a chemically synthesized double stranded DNA template and T7 polymerase. The RNA product was purified on a 15% denaturing polyacrylamide gel and

recovered by electroelution and ethanol precipitation. The RNA was exchanged into pH 6.4 NMR buffer containing 15 mM sodium phosphate, 25 mM sodium chloride, and 0.1 mM EDTA. Each TAR sample had 22 Watson-Crick base pairs added to the terminal end of domain I, for the purposes of obtaining relaxation parameters (Figure 1a). In one sample, domain I was elongated with adenine-uracil (AU) base pairs and transcribed with 13C- and 15N-labeled guanine and cytosine. The other sample was elongated with guanine-cytosine (GC) base pairs and transcribed with labeled adenine and uracil. This labeling strategy was employed so as to avoid significant peak overlap due to the additional residues in the elongated domain. An overlay of heteronuclear single quantum coherence (HSQC) spectra from the elongated and nonelongated constructs show very few differences, excepting for those residues directly adjacent to the residues added in elongation, indicating that there are very few structural differences between the constructs (see Figure 1d). 15 N relaxation studies were performed on the elongated TAR samples at a field strength of 600 MHz on an Avance Bruker spectrometer equipped with a triple resonance 5 mm cryogenic probe. Longitudinal (R1) and transverse (R2(CPMG)) relaxation rates and NOEs were measured at 298 K using standard 1D and 2D R1, R2, and NOE experiments on the AU- and GClabeled samples, respectively. In the R1 experiment, the imino 1 H spins were decoupled during the relaxation delay using 1250 µs iburp2 pulses centered on the imino signals. A recycling delay of 1.9 s was used, and relaxation delays of 60, 120, 240, 480, 640, 800, and 1200 ms were used, repeating experiments with delays of 120 and 800 ms for the purposes of error analysis. In the R2 experiment, a [0013]N phase cycling scheme was employed to suppress off-resonance effects.30 As in the R1 experiments, the imino proton spins were decoupled during the relaxation delay using 1250 µs ipurb2 pulses. The interpulse delay for the CPMG scheme was 1300 µs. A recycling delay of 1.9 s was used, and relaxation delays of 6.2, 12.4, 24.8, 37.2, 49.6, 62.0, and 74.4 ms were implmeneted, with repeated experiments performed for delays of 12.4 and 62.0 ms for the purposes of error analysis. Rates R1 and R2 were obtained through fitting the relaxation data to exponential functions (with an offset of 0) using the program Origin 7.0 (OriginLab Corp.). The R2 values were corrected for off-resonance effects as described in the work of Yip et al.30 Errors in R1 and R2 were taken to be the fitting errors as determined using Origin. NOE errors were obtained from NMRDraw and were taken as the square root the sum of the squares of the errors in the two peak heights. Data was obtained for N1H1 or N3H3 bonds in residues G17, G18, G21, and U42 in domain I and residues G28, G36, and U38 in domain II. The peaks for G26 and G43 could not be resolved, and thus, data for these residues were not obtained. Additionally, the lack of a hydrogen bond between A22 and U40 precluded the ability to obtain data for U40. D. Model-Free Analysis of Experimental Data. The relaxation parameters were interpreted according to the modelfree formalism yielding parameters describing the amplitude (S2) and time scale (τ) of bond vector motion. The program ModelFree provided by the Palmer laboratory (http://www. palmer.hs.columbia.edu/31) was used to best fit the relaxation data to dynamical parameters. The input structure for all calculations was built using ideal A-form models of the two helices and the loop from structure 3 of the NMR solved ensemble of structures of TAR,32 with the average global orientation of the helices obtained from a residual dipolar coupling (RDC) analysis (data not shown). The program

932

J. Phys. Chem. B, Vol. 114, No. 2, 2010

HYDRONMR33 was utilized to estimate the rotational diffusion tensor parameters based on the hydrodynamic shape of the structure, with input parameters AER ) 3.2, NSIG ) 6, SIGMIN ) 1.4, SIGMAX )2.4, T ) 298.15 K, and a solvent viscosity of 0.008906. The resulting diffusion tensor parameters were Dzz ) 17.6 MHz, Dyy ) 3.53 MHz, and Dxx ) 3.47 MHz. The tensor parameters obtained from HYDRONMR were then further refined using ModelFree with the input structure rotated into the determined hydrodynamic principal axis system (PAS). For the purposes of the ModelFree calculations, the determined errors in R1 and R2 were doubled and the NOE error was set to 8%. A static bond length of 1.01 Å was used,21,65 and ∆σ was set to -130 and -100 ppm for N1 and N3, respectively. All calculations utilized 300 randomly distributed data sets generated by Monte Carlo simulations. To refine the diffusion parameters, the rotational correlation time and polar angles θ and φ (defining the orientation of the z-component of the diffusion tensor) were varied from 15 to 24 ns and from 0 to 20° and 0 to 360°, respectively, with Dratio set to the HYDRONMR predicted value (Dratio ) 5.0). This procedure was performed with data for domain I from the GC-labeled construct utilizing the Powell minimization algorithm, yielding τm ) 18.7 ns, θ ) 18.1°, and φ ) 151.2°. These simulations were repeated with 10% variations in Dratio with very similar results. Using a structure in the GC-labeled diffusion tensor frame τm was refined for the AU-labeled domain I data using the simulated annealing protocol with “grid 100 50 0.9”, yielding τm ) 20.7 ns. Utilizing these parameters, the procedure outlined by Mandel et al.31 was followed for model selection for individual bond vectors. Model 1 (S2) was chosen for G18 and U42, model 2 (S2 and τ) was chosen for G17 and G21, and model 5 (Ss2, S2f , and τs) was chosen for G28, G36, and U38. These parameters were then refined using the Powell minimization and simulated annealing algorithms for the GC- and AUlabeled data, respectively. E. MD Simulation. A 65-ns MD simulation of wild-type HIV-1 TAR (see Figure 1a) was performed using the CHARMM force field with the parameter set 27.34 Details of the simulation protocol and the first 20 ns of the resultant trajectory have been reported previously.24,28,64 In what follows, we give a brief description of the setup for the presently reported trajectory. Starting coordinates were obtained from structure 3 of the family of free TAR NMR structures (PDB 1ANR).35 This structure was chosen as it yields the best agreement with previously measured residual dipolar couplings (RDCs).36 The RNA was charge-neutralized using sodium counterions and solvated in a 35 Å sphere of TIP3P water,37 which allowed for >9 Å distance between the surface of the sphere and all RNA atoms. A stochastic boundary potential on the water molecules was used.38 This system was minimized and heated to 300 K, while harmonically constraining the heavy atoms of the RNA with a force constant of 62 kcal/(mol Å2) for 100 ps, after which the constraints were removed and the system equilibrated for 1 ns. A Nose´-Hoover thermostat39,40 was used to maintain a constant temperature of 300 K throughout the simulation, with a 1 fs time step and a bath coupling constant of 50 1/ps. F. Calculated Parameters. Relaxation parameters R1, R2, and NOE were calculated from the MD trajectory for N1H1/ N3H3 vectors for residues G17, G18, G21, and U42 in domain I and residues G28, G36, and U38 in domain II. In order to be able to rigorously compare the calculated and experimental parameters, it was necessary to match the experimental diffusion tensor frame. Since domain I of TAR was elongated in the experiments, it dominates overall tumbling and is the reference

Musselman et al. frame for all experimental parameters. Thus, to mimic this reference frame, the MD relaxation parameters were calculated with the domain I heavy atoms as a reference for overlaying trajectory snapshots, excluding A22-U40 as this base pair is not hydrogen bonded.32 This referencing strategy anchors the MD reference frame to domain I and should render the computed and experimental parameters comparable. The overall rotational tumbling was modeled by an axially symmetric diffusion tensor, which is supported by the HYDRONMR calculation of the tensor parameters, which found Dxx ∼ Dyy. Values for the diffusion tensor were set equal to those derived from the model free analysis of the experimental data, where the principle components of the diffusion tensor, D|| and D⊥, were found to be 19.1 and 3.8 MHz, respectively, for the guanine data and 17.3 and 3.5 MHz, respectively, for the uracil data, with a magnetic field of 14.1 T and ∆σ of -100 and -130 ppm for N3 and N1, respectively. The domain I axis is, to a good approximation, collinear with the principal axis of the experimental diffusion tensor, thus the average angle of each bond vector with respect to the principal axis of the diffusion tensor was approximated by the average angle between the bond vector and the helix axis of domain I, which was determined using the program CURVES.41 Values for 〈r3〉 were calculated as an average over the MD trajectory and were within 0.005 Å of the static bond lengths used in the experimental analysis. All correlation functions for use in calculating the spectral density functions were computed from 1 to 65 ns of the trajectory with tmax ) 6.5 ns (excepting those discussed in section III.D.). For converged correlation functions, Ci(∞) was calculated using the equilibrium expression for S2 (see eq 6). However, for those functions which were not converged during the simulation, Ci(∞) was approximated as the tail value of the correlation function, defined as the average value over the last 500 ps in the C(t) dependence. Internal correlation functions for domain II bond vectors were calculated using two reference frames. The domain I reference frame yields functions which contain information on both the local motions of the bond vectors within domain II and any motions induced by the fluctuation of domain II with respect to domain I. Additionally, to obtain information on only the local motions within the domain correlation functions were calculated using domain II as the reference frame. Here the terminal residues and those neighboring the bulge and loop were excluded from the overlay, as they are expected to experience significant dynamics. Assuming that the local and domain motions are separable, dividing the correlation functions calculated for bond vectors in domain II when using domain I as the reference (CI(t)) by the correlation functions obtained using domain II as the reference frame (CII(t)) yields correlation functions containing information on only the motions of the bond vector induced by the slow, global motions of domain II with respect to domain I,

Cs(t) ) CI(t)/CII(t)

(7)

Order parameters describing the local motions of bond vectors were calculated using eq 6 (Seq2) from 1 to 65 ns of the MD trajectory as well calculated from the tail values Stail2 of the correlation functions (see above). To test for the convergence of the correlation functions Seq2 values were compared with Stail2 values, the functions were considered converged if |Stail2 - Seq2| < 0.005. All correlation functions were parametrized by fitting to single, double, and triple exponential functions using nonlinear least-squares fitting of the form

J. Phys. Chem. B, Vol. 114, No. 2, 2010 933 n

Ci(t) ) c0 +

∑ cje-t/τ

j

(8)

j)1

where cj, j ) 0, n, are constants, n ) 1 corresponds to a single exponential, n ) 2 to a double exponential, and n ) 3 to a triple exponential fit, and where τj is the correlation time constant. The quality of fit was determined by analysis of the χ2/dof and R2 values obtained for each fit.42 Errors in the correlation functions were approximated as43

σC(t) ) √2τint /T(1 - C(t)/C(0))

(9)

where T is the upper time limit of the trajectory (i.e., T ) 64 ns) and τint is the integrated correlation time computed as

τint )

1 Ci(0) - Ci(∞)

∫0∞ (Ci(t) - Ci(∞)) dt

(10)

Order parameter errors were approximated to be equal to the error in the last point of the correlation function. Errors in the relaxation parameters were estimated from the standard deviation in a set of 1000 computed relaxation parameters in which Gaussian errors were added to each time point of the internal correlation functions with a standard deviation of σC(t). Errors in the correlation time constants were obtained from the fitting errors of correlation functions in which each time point had added to it error with a standard deviation of σC(t). In these fittings, the coefficients in the exponential functions were set to those obtained when fitting the functions without any added error. III. Results and Discussion In an elongated-domain NMR experiment, as described above, one helical domain of an RNA is elongated using a stretch of Watson-Crick base-pairs. This slows down overall rotational diffusion, thus decoupling it from internal motions that occur on time scales approaching the overall tumbling of the nonelongated construct. Additionally, as the elongated domain significantly dominates the rotational diffusion of the RNA, it becomes the reference frame for all measured parameters. Thus, dynamical parameters will report only on the local motions of bond vectors in the elongated domain but, for any additional domains, will report on both local and interdomain motions. In this way, elongation can reveal domain motions that are hidden in the nonelongated constructs due to coupling with the overall rotational diffusion. A similar frame of reference can easily be obtained in the analysis of an MD trajectory. To accomplish this, the domain which has been elongated in the experiment (though not necessarily in the simulation) is used as the reference frame for overlaying each snapshot from the MD trajectory by minimization of the root-mean-square atomic displacements of the heavy atoms. As in the experimental approach, this results in any calculated parameters being inherently referenced to this domain, rendering them comparable to the experimental values. We have applied this strategy to HIV-1 TAR RNA (see Figure 1a), for which we obtained relaxation constants R1, R2, and NOE from an elongated-domain experiment and from a 65 ns MD trajectory. Presented below is a discussion of the cross-validation of the relaxation and associated model-free parameters, and an analysis of what they reveal about the dynamics of TAR.

A. Domain I: Reference Domain. Relaxation parameters (R1, R2, and NOE) obtained experimentally and from the 65 ns MD simulation are shown in Figure 2. These parameters are shown as a function of residue and separated by domain. As explained above, dynamical parameters in domain I will report only on local motions, which are expected to be on the picosecond time scale. Since this is significantly shorter than the rate of overall tumbling in the NMR experiment (τm ≈ 20 ns) and than the length of the MD trajectory (T ) 65 ns), these motions should be easily resolved by both techniques. Looking at the values for the reference, domain I (residues 17-22 and 40-45; see Figure 2), very good agreement is observed between the calculated (green) and experimental values (gray), with R1 values differing on average by only 3.1%, R2 values by 3.9%, and NOE values by 13.4%. This good agreement between the experimental and computed values in the reference domain shows that the MD simulation is reliably reproducing the internal dynamics in the domain and also demonstrates the validity of the referencing strategy. The local motions in domain I were also characterized through experimental and computed model-free parameters. Shown in Figure 3 are the order parameters and correlation times for local motion, again as a function of residue and separated by domain. Excellent agreement between the experimental and computed S2 values is observed for domain I, indicating that the amplitude of bond vector motions gauged by the two techniques is approximately the same, and the average values of 0.88 and 0.85 obtained from NMR and MD, respectively, signify a level of dynamics expected for a standard A-form helix.14,28 For the associated time scales, the comparison becomes more complicated. Model-free analysis of NMR data typically relies on either a single exponential10 (models 1 and 2 in ModelFree) or, in the extended formalism, a double exponential fit11 (model 5) of the correlation function, with further parametrization not possible given the limited data available. For all domain I bond vectors either model 1 or 2 was chosen with resulting time scales of